xref: /llvm-project/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll (revision 76c22b18eafd2156568d72e9df2ff7bd3457888a)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
3; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s
4; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
5
6define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 {
7; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict:
8; GCN:       ; %bb.0:
9; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
10; GCN-NEXT:    v_add_f32_e32 v0, v0, v1
11; GCN-NEXT:    s_setpc_b64 s[30:31]
12;
13; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict:
14; GFX10PLUS:       ; %bb.0:
15; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
16; GFX10PLUS-NEXT:    v_add_f32_e32 v0, v0, v1
17; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
18  %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
19  ret float %val
20}
21
22define float @v_constained_fadd_f32_fpexcept_ignore(float %x, float %y) #0 {
23; GCN-LABEL: v_constained_fadd_f32_fpexcept_ignore:
24; GCN:       ; %bb.0:
25; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
26; GCN-NEXT:    v_add_f32_e32 v0, v0, v1
27; GCN-NEXT:    s_setpc_b64 s[30:31]
28;
29; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_ignore:
30; GFX10PLUS:       ; %bb.0:
31; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
32; GFX10PLUS-NEXT:    v_add_f32_e32 v0, v0, v1
33; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
34  %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
35  ret float %val
36}
37
38define float @v_constained_fadd_f32_fpexcept_maytrap(float %x, float %y) #0 {
39; GCN-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
40; GCN:       ; %bb.0:
41; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
42; GCN-NEXT:    v_add_f32_e32 v0, v0, v1
43; GCN-NEXT:    s_setpc_b64 s[30:31]
44;
45; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
46; GFX10PLUS:       ; %bb.0:
47; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
48; GFX10PLUS-NEXT:    v_add_f32_e32 v0, v0, v1
49; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
50  %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
51  ret float %val
52}
53
54define <2 x float> @v_constained_fadd_v2f32_fpexcept_strict(<2 x float> %x, <2 x float> %y) #0 {
55; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
56; GCN:       ; %bb.0:
57; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
58; GCN-NEXT:    v_add_f32_e32 v0, v0, v2
59; GCN-NEXT:    v_add_f32_e32 v1, v1, v3
60; GCN-NEXT:    s_setpc_b64 s[30:31]
61;
62; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
63; GFX10:       ; %bb.0:
64; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
65; GFX10-NEXT:    v_add_f32_e32 v0, v0, v2
66; GFX10-NEXT:    v_add_f32_e32 v1, v1, v3
67; GFX10-NEXT:    s_setpc_b64 s[30:31]
68;
69; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
70; GFX11:       ; %bb.0:
71; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
72; GFX11-NEXT:    v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
73; GFX11-NEXT:    s_setpc_b64 s[30:31]
74  %val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
75  ret <2 x float> %val
76}
77
78define <2 x float> @v_constained_fadd_v2f32_fpexcept_ignore(<2 x float> %x, <2 x float> %y) #0 {
79; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
80; GCN:       ; %bb.0:
81; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
82; GCN-NEXT:    v_add_f32_e32 v0, v0, v2
83; GCN-NEXT:    v_add_f32_e32 v1, v1, v3
84; GCN-NEXT:    s_setpc_b64 s[30:31]
85;
86; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
87; GFX10:       ; %bb.0:
88; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
89; GFX10-NEXT:    v_add_f32_e32 v0, v0, v2
90; GFX10-NEXT:    v_add_f32_e32 v1, v1, v3
91; GFX10-NEXT:    s_setpc_b64 s[30:31]
92;
93; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
94; GFX11:       ; %bb.0:
95; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
96; GFX11-NEXT:    v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
97; GFX11-NEXT:    s_setpc_b64 s[30:31]
98  %val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
99  ret <2 x float> %val
100}
101
102define <2 x float> @v_constained_fadd_v2f32_fpexcept_maytrap(<2 x float> %x, <2 x float> %y) #0 {
103; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
104; GCN:       ; %bb.0:
105; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
106; GCN-NEXT:    v_add_f32_e32 v0, v0, v2
107; GCN-NEXT:    v_add_f32_e32 v1, v1, v3
108; GCN-NEXT:    s_setpc_b64 s[30:31]
109;
110; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
111; GFX10:       ; %bb.0:
112; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
113; GFX10-NEXT:    v_add_f32_e32 v0, v0, v2
114; GFX10-NEXT:    v_add_f32_e32 v1, v1, v3
115; GFX10-NEXT:    s_setpc_b64 s[30:31]
116;
117; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
118; GFX11:       ; %bb.0:
119; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
120; GFX11-NEXT:    v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
121; GFX11-NEXT:    s_setpc_b64 s[30:31]
122  %val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
123  ret <2 x float> %val
124}
125
126define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x float> %y) #0 {
127; GCN-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
128; GCN:       ; %bb.0:
129; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
130; GCN-NEXT:    v_add_f32_e32 v0, v0, v3
131; GCN-NEXT:    v_add_f32_e32 v1, v1, v4
132; GCN-NEXT:    v_add_f32_e32 v2, v2, v5
133; GCN-NEXT:    s_setpc_b64 s[30:31]
134;
135; GFX10-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
136; GFX10:       ; %bb.0:
137; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
138; GFX10-NEXT:    v_add_f32_e32 v0, v0, v3
139; GFX10-NEXT:    v_add_f32_e32 v1, v1, v4
140; GFX10-NEXT:    v_add_f32_e32 v2, v2, v5
141; GFX10-NEXT:    s_setpc_b64 s[30:31]
142;
143; GFX11-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
144; GFX11:       ; %bb.0:
145; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
146; GFX11-NEXT:    v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
147; GFX11-NEXT:    v_add_f32_e32 v2, v2, v5
148; GFX11-NEXT:    s_setpc_b64 s[30:31]
149  %val = call <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
150  ret <3 x float> %val
151}
152
153define amdgpu_ps float @s_constained_fadd_f32_fpexcept_strict(float inreg %x, float inreg %y) #0 {
154; GCN-LABEL: s_constained_fadd_f32_fpexcept_strict:
155; GCN:       ; %bb.0:
156; GCN-NEXT:    v_mov_b32_e32 v0, s3
157; GCN-NEXT:    v_add_f32_e32 v0, s2, v0
158; GCN-NEXT:    ; return to shader part epilog
159;
160; GFX10PLUS-LABEL: s_constained_fadd_f32_fpexcept_strict:
161; GFX10PLUS:       ; %bb.0:
162; GFX10PLUS-NEXT:    v_add_f32_e64 v0, s2, s3
163; GFX10PLUS-NEXT:    ; return to shader part epilog
164  %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
165  ret float %val
166}
167
168define float @v_constained_fadd_f32_fpexcept_strict_fabs_lhs(float %x, float %y) #0 {
169; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
170; GCN:       ; %bb.0:
171; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
172; GCN-NEXT:    v_add_f32_e64 v0, |v0|, v1
173; GCN-NEXT:    s_setpc_b64 s[30:31]
174;
175; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
176; GFX10PLUS:       ; %bb.0:
177; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
178; GFX10PLUS-NEXT:    v_add_f32_e64 v0, |v0|, v1
179; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
180  %fabs.x = call float @llvm.fabs.f32(float %x) #0
181  %val = call float @llvm.experimental.constrained.fadd.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
182  ret float %val
183}
184
185define float @v_constained_fadd_f32_fpexcept_strict_fabs_rhs(float %x, float %y) #0 {
186; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
187; GCN:       ; %bb.0:
188; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
189; GCN-NEXT:    v_add_f32_e64 v0, v0, |v1|
190; GCN-NEXT:    s_setpc_b64 s[30:31]
191;
192; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
193; GFX10PLUS:       ; %bb.0:
194; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
195; GFX10PLUS-NEXT:    v_add_f32_e64 v0, v0, |v1|
196; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
197  %fabs.y = call float @llvm.fabs.f32(float %y) #0
198  %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
199  ret float %val
200}
201
202define float @v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs(float %x, float %y) #0 {
203; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
204; GCN:       ; %bb.0:
205; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
206; GCN-NEXT:    v_sub_f32_e64 v0, v1, |v0|
207; GCN-NEXT:    s_setpc_b64 s[30:31]
208;
209; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
210; GFX10PLUS:       ; %bb.0:
211; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
212; GFX10PLUS-NEXT:    v_sub_f32_e64 v0, v1, |v0|
213; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
214  %fabs.x = call float @llvm.fabs.f32(float %x) #0
215  %neg.fabs.x = fneg float %fabs.x
216  %val = call float @llvm.experimental.constrained.fadd.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
217  ret float %val
218}
219
220declare float @llvm.fabs.f32(float)
221declare float @llvm.experimental.constrained.fadd.f32(float, float, metadata, metadata)
222declare <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float>, <2 x float>, metadata, metadata)
223declare <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float>, <3 x float>, metadata, metadata)
224
225attributes #0 = { strictfp }
226